Spatial Sign Preprocessing:  A Simple Way To Impart Moderate Robustness to Multivariate Estimators

Serneels, Sven; Nolf, Evert De; Espen, Pierre J. Van

doi:10.1021/ci050498u

Cited by 47 publications

(25 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Next, we employed global normalization in which the vector of RNA abundance (within each category) was divided by the norm of the vector ( Serneels et al, 2006 ). This method also imparted robustness to outliers.…”

Section: Methodsmentioning

confidence: 99%

Validation of a Salivary RNA Test for Childhood Autism Spectrum Disorder

Hicks

Rajan²,

Wagner

et al. 2018

Front. Genet.

View full text Add to dashboard Cite

Background: The diagnosis of autism spectrum disorder (ASD) relies on behavioral assessment. Efforts to define biomarkers of ASD have not resulted in an objective, reliable test. Studies of RNA levels in ASD have demonstrated potential utility, but have been limited by a focus on single RNA types, small sample sizes, and lack of developmental delay controls. We hypothesized that a saliva-based poly-“omic” RNA panel could objectively distinguish children with ASD from their neurotypical peers and children with non-ASD developmental delay.Methods: This multi-center cross-sectional study included 456 children, ages 19–83 months. Children were either neurotypical (n = 134) or had a diagnosis of ASD (n = 238), or non-ASD developmental delay (n = 84). Comprehensive human and microbial RNA abundance was measured in the saliva of all participants using unbiased next generation sequencing. Prior to analysis, the sample was randomly divided into a training set (82% of subjects) and an independent validation test set (18% of subjects). The training set was used to develop an RNA-based algorithm that distinguished ASD and non-ASD children. The validation set was not used in model development (feature selection or training) but served only to validate empirical accuracy.Results: In the training set (n = 372; mean age 51 months; 75% male; 51% ASD), a set of 32 RNA features (controlled for demographic and medical characteristics), identified ASD status with a cross-validated area under the curve (AUC) of 0.87 (95% CI: 0.86–0.88). In the completely separate validation test set (n = 84; mean age 50 months; 85% male; 60% ASD), the algorithm maintained an AUC of 0.88 (82% sensitivity and 88% specificity). Notably, the RNA features were implicated in physiologic processes related to ASD (axon guidance, neurotrophic signaling).Conclusion: Salivary poly-omic RNA measurement represents a novel, non-invasive approach that can accurately identify children with ASD. This technology could improve the specificity of referrals for ASD evaluation or provide objective support for ASD diagnoses.

show abstract

Section: Methodsmentioning

confidence: 99%

Validation of a Salivary RNA Test for Childhood Autism Spectrum Disorder

Hicks

Rajan²,

Wagner

et al. 2018

Front. Genet.

View full text Add to dashboard Cite

show abstract

“…The third, RSIMPLS, is the algorithm proposed by Hubert and Vanden Branden [7]. The last two robust procedures are PRM, the partial robust M-estimator proposed by Serneels, Croux, Filzmoser and Van Espen [14], and the one called SS-PLS (see Reference [15]) which is a robustification of the standard PLS algorithm obtained by spatial sign preprocessing in combination with PLS regression. First, we test the algorithms when several types of outliers are present in the data.…”

Section: Numerical Resultsmentioning

confidence: 99%

“…In order to check the efficiency of the algorithms in the case of having non normal error in the original non contaminated model, we carried out the following experiment presented in Reference [15]. We work with the initial model…”

Section: Monte Carlo Experiments With Noise Modelsmentioning

confidence: 99%

A robust partial least squares regression method with applications

González

Peña

Romera

2008

Journal of Chemometrics

View full text Add to dashboard Cite

Partial least squares (PLS) regression is a linear regression technique developed to relate many regressors to one or several response variables. Robust methods are introduced to reduce or remove the effect of outlying data points. In this paper, we show that if the sample covariance matrix is properly robustified further robustification of the linear regression steps of the PLS algorithm becomes unnecessary. The robust estimate of the covariance matrix is computed by searching for outliers in univariate projections of the data on a combination of random directions (Stahel-Donoho) and specific directions obtained by maximizing and minimizing the kurtosis coefficient of the projected data, as proposed by Peña and Prieto [1]. It is shown that this procedure is fast to apply and provides better results than other methods proposed in the literature. Its performance is illustrated by Monte Carlo and by an example, where the algorithm is able to show features of the data which were undetected by previous methods.

show abstract

“…Spatial sign preprocessing is a relatively simple data transformation that facilitates the construction of robust PLS [42]. Similar to elliptical PCA, data are projected onto a sphere of unit radius and a centre coinciding with the robust data centre.…”

Section: Spatial Sign Preprocessing and Robust Plsmentioning

confidence: 99%